Sequential Bayesian Network Structure Learning

Sachini Piyoni Ekanayake, Daphney-Stavroula Zois

Abstract: In many real-world applications, e.g., medical diagnosis, behavioral analysis, Bayesian networks are used to describe relationships between variables. In this context, a very important task is learning the underlying structure of such networks. However, this constitutes an NP-hard problem. In this paper, we propose an approach to speed-up the structure learning process without compromising accuracy, assuming a given set of candidate network structures. Specifically, the proposed method sequentially evaluates variable relationships until it reaches a specific decision regarding the underlying Bayesian network. The performance of the proposed approach is illustrated on two standard Bayesian networks and compared with existing methods.